A reliable identification of seismic precursors requires a two-step statistical approach: a retrospective ''learning'' step to establish the candidate precursors on the basis of ad hoc chosen laws of the game, and a ''validation'' step which evaluates candidate precursors on an independent data set according to the previously established laws of the game. No published work has so far provided a sound support to the candidacy of the VAN signals as earthquake precursors. The problem is complicated by a chronic fuzziness in the text of VAN's predictions, which often allows different interpretations, and by the absence of any clear statement by VAN of the laws of the game. ¿ American Geophysical Union 1996